Research On Multi-target Trajectory Extraction And Pattern Mining Algorithm Based On Passive Locating Data

Passive locating is the technique of obtaining target position information by receiving radiation source signal through passive sensors.It has the advantages of invisibility,long detection distance,strong adaptability etc.,and the ability of positioning multiple targets fast and accurately.By analyzing a large number of spatiotemporal location data obtained by passive location,the target frequent active regions and action trajectories can be obtained.If the time periodicity and historical behavior pattern of the target activity are used in real-time positioning and tracking,the abnormal activities of the target can be found by matching the typical feature movement patterns of the target.Therefore,it is theoretically and practically valuable to study the passive location data mining technology and find the targets' behavior pattern.Primarily,focusing on the multi-target and variable locating period characteristics of passive locating data,the three-stage iterative associating trajectory extraction algorithm based on improved track initiation model is proposed.Then,the technology of spatio-temporal data mining is introduced into the passive location data domain.Clustering algorithm is used to extract the similarity and abnormality in the data.In the end,the trajectory data mining and early warning platform are designed and implemented based on the key technologies proposed above.The main contributions and innovations of this paper are described as follow:1.The three-stage iterative correlation trajectory extraction technology based on improved track initiation model is proposed.The dynamic extrapolation rule based on Singer motion model is designed to reduce the extrapolation error and meet the requirements for different trajectory extraction of historical data and real-time data.In order to judge the time interval of measurement,the frame of the passive locating data trajectories extraction process is put forward.The weighted scoring method and the steering angle constraint are used to improve the trajectory quality management algorithm.Finally,the three stage iterative Association trajectory extraction algorithm is formed.The simulation results show that,compared with the traditional method,the algorithm in the passive direction finding location scene,when the location point loss rate is high,the trajectory extraction accuracy is far higher than the traditional common track starting model extraction method.2.,A trajectory pattern mining algorithm based on trajectory structure and the longest common subsequence(TS-LCSS)clustering is proposed.In order to recognize both track posture and structure,the idea of density based clustering algorithm(DBSCAN)and two step clustering is adopted to first carry out rough clustering based on the longest common subsequence(LCSS),and then divide the trajectories in the rough clustering to carry on the sub-trajectory clustering based on the similarity of the trajectory structure(TS).The clustering accuracy of track structure and trajectory situation is improved.In order to solve the problem of poor real-time performance of matching between trajectory and a large number of historical trajectories,the history representative trajectory algorithm is proposed.It greatly reduces the matching time while ensuring the accuracy of track pattern matching.In view of the problem that the traditional DBSCAN needs to determine the parameters through the experiment,the DBSCAN parameter adaptive calculation method is designed.The traditional LCSS's Euclidean distance measurement cannot fully reflect the trajectory structure.The traditional LCSS distance measurement of points is improved to the sub-trajectories.And the LCSS fast calculation method is added.The traditional trajectory segmentation algorithm is prone to error segmentation due to the interference of noise points,so the adjacent trajectory angle segmentation strategy is changed to the integral trajectory angle segmentation.The simulation results show that compared with the traditional angle clustering and the path structure clustering algorithm,the TS-LCSS algorithm is more comprehensive in the analysis of the trajectory.The clustering results have more practical significance and improve the accuracy of clustering.Compared with the method of one-by-one matching,the representative trajectory matching algorithm is accurate and real-time.3.Integrating the above algorithm functions,a trajectory data mining and warning system is designed and accomplished.The software system sets scene layout,track data mining algorithm research and real-time tracking warning in one.It has the advantages of automation,visualization and extensibility.The system includes three parts: passive location database,algorithm function module and interactive interface,which can excavate historical passive location data,carry out real-time trajectory extraction and position prediction,target motion pattern discovery and early warning function.The performance test results show that trajectory extraction,clustering,pattern matching and early warning precision is high and track pattern matching is real-time.